1 //===- DWARFDebugLine.cpp -------------------------------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 9 #include "llvm/DebugInfo/DWARF/DWARFDebugLine.h" 10 #include "llvm/ADT/Optional.h" 11 #include "llvm/ADT/SmallString.h" 12 #include "llvm/ADT/SmallVector.h" 13 #include "llvm/ADT/StringRef.h" 14 #include "llvm/BinaryFormat/Dwarf.h" 15 #include "llvm/DebugInfo/DWARF/DWARFDataExtractor.h" 16 #include "llvm/DebugInfo/DWARF/DWARFDie.h" 17 #include "llvm/DebugInfo/DWARF/DWARFFormValue.h" 18 #include "llvm/Support/Errc.h" 19 #include "llvm/Support/Format.h" 20 #include "llvm/Support/FormatVariadic.h" 21 #include "llvm/Support/raw_ostream.h" 22 #include <algorithm> 23 #include <cassert> 24 #include <cinttypes> 25 #include <cstdint> 26 #include <cstdio> 27 #include <utility> 28 29 using namespace llvm; 30 using namespace dwarf; 31 32 namespace llvm { 33 class DwarfContext; 34 } 35 36 using FileLineInfoKind = DILineInfoSpecifier::FileLineInfoKind; 37 38 namespace { 39 40 struct ContentDescriptor { 41 dwarf::LineNumberEntryFormat Type; 42 dwarf::Form Form; 43 }; 44 45 using ContentDescriptors = SmallVector<ContentDescriptor, 4>; 46 47 } // end anonymous namespace 48 49 static bool versionIsSupported(uint16_t Version) { 50 return Version >= 2 && Version <= 5; 51 } 52 53 void DWARFDebugLine::ContentTypeTracker::trackContentType( 54 dwarf::LineNumberEntryFormat ContentType) { 55 switch (ContentType) { 56 case dwarf::DW_LNCT_timestamp: 57 HasModTime = true; 58 break; 59 case dwarf::DW_LNCT_size: 60 HasLength = true; 61 break; 62 case dwarf::DW_LNCT_MD5: 63 HasMD5 = true; 64 break; 65 case dwarf::DW_LNCT_LLVM_source: 66 HasSource = true; 67 break; 68 default: 69 // We only care about values we consider optional, and new values may be 70 // added in the vendor extension range, so we do not match exhaustively. 71 break; 72 } 73 } 74 75 DWARFDebugLine::Prologue::Prologue() { clear(); } 76 77 bool DWARFDebugLine::Prologue::hasFileAtIndex(uint64_t FileIndex) const { 78 uint16_t DwarfVersion = getVersion(); 79 assert(DwarfVersion != 0 && 80 "line table prologue has no dwarf version information"); 81 if (DwarfVersion >= 5) 82 return FileIndex < FileNames.size(); 83 return FileIndex != 0 && FileIndex <= FileNames.size(); 84 } 85 86 Optional<uint64_t> DWARFDebugLine::Prologue::getLastValidFileIndex() const { 87 if (FileNames.empty()) 88 return None; 89 uint16_t DwarfVersion = getVersion(); 90 assert(DwarfVersion != 0 && 91 "line table prologue has no dwarf version information"); 92 // In DWARF v5 the file names are 0-indexed. 93 if (DwarfVersion >= 5) 94 return FileNames.size() - 1; 95 return FileNames.size(); 96 } 97 98 const llvm::DWARFDebugLine::FileNameEntry & 99 DWARFDebugLine::Prologue::getFileNameEntry(uint64_t Index) const { 100 uint16_t DwarfVersion = getVersion(); 101 assert(DwarfVersion != 0 && 102 "line table prologue has no dwarf version information"); 103 // In DWARF v5 the file names are 0-indexed. 104 if (DwarfVersion >= 5) 105 return FileNames[Index]; 106 return FileNames[Index - 1]; 107 } 108 109 void DWARFDebugLine::Prologue::clear() { 110 TotalLength = PrologueLength = 0; 111 SegSelectorSize = 0; 112 MinInstLength = MaxOpsPerInst = DefaultIsStmt = LineBase = LineRange = 0; 113 OpcodeBase = 0; 114 FormParams = dwarf::FormParams({0, 0, DWARF32}); 115 ContentTypes = ContentTypeTracker(); 116 StandardOpcodeLengths.clear(); 117 IncludeDirectories.clear(); 118 FileNames.clear(); 119 } 120 121 void DWARFDebugLine::Prologue::dump(raw_ostream &OS, 122 DIDumpOptions DumpOptions) const { 123 if (!totalLengthIsValid()) 124 return; 125 int OffsetDumpWidth = 2 * dwarf::getDwarfOffsetByteSize(FormParams.Format); 126 OS << "Line table prologue:\n" 127 << format(" total_length: 0x%0*" PRIx64 "\n", OffsetDumpWidth, 128 TotalLength) 129 << " format: " << dwarf::FormatString(FormParams.Format) << "\n" 130 << format(" version: %u\n", getVersion()); 131 if (!versionIsSupported(getVersion())) 132 return; 133 if (getVersion() >= 5) 134 OS << format(" address_size: %u\n", getAddressSize()) 135 << format(" seg_select_size: %u\n", SegSelectorSize); 136 OS << format(" prologue_length: 0x%0*" PRIx64 "\n", OffsetDumpWidth, 137 PrologueLength) 138 << format(" min_inst_length: %u\n", MinInstLength) 139 << format(getVersion() >= 4 ? "max_ops_per_inst: %u\n" : "", MaxOpsPerInst) 140 << format(" default_is_stmt: %u\n", DefaultIsStmt) 141 << format(" line_base: %i\n", LineBase) 142 << format(" line_range: %u\n", LineRange) 143 << format(" opcode_base: %u\n", OpcodeBase); 144 145 for (uint32_t I = 0; I != StandardOpcodeLengths.size(); ++I) 146 OS << formatv("standard_opcode_lengths[{0}] = {1}\n", 147 static_cast<dwarf::LineNumberOps>(I + 1), 148 StandardOpcodeLengths[I]); 149 150 if (!IncludeDirectories.empty()) { 151 // DWARF v5 starts directory indexes at 0. 152 uint32_t DirBase = getVersion() >= 5 ? 0 : 1; 153 for (uint32_t I = 0; I != IncludeDirectories.size(); ++I) { 154 OS << format("include_directories[%3u] = ", I + DirBase); 155 IncludeDirectories[I].dump(OS, DumpOptions); 156 OS << '\n'; 157 } 158 } 159 160 if (!FileNames.empty()) { 161 // DWARF v5 starts file indexes at 0. 162 uint32_t FileBase = getVersion() >= 5 ? 0 : 1; 163 for (uint32_t I = 0; I != FileNames.size(); ++I) { 164 const FileNameEntry &FileEntry = FileNames[I]; 165 OS << format("file_names[%3u]:\n", I + FileBase); 166 OS << " name: "; 167 FileEntry.Name.dump(OS, DumpOptions); 168 OS << '\n' 169 << format(" dir_index: %" PRIu64 "\n", FileEntry.DirIdx); 170 if (ContentTypes.HasMD5) 171 OS << " md5_checksum: " << FileEntry.Checksum.digest() << '\n'; 172 if (ContentTypes.HasModTime) 173 OS << format(" mod_time: 0x%8.8" PRIx64 "\n", FileEntry.ModTime); 174 if (ContentTypes.HasLength) 175 OS << format(" length: 0x%8.8" PRIx64 "\n", FileEntry.Length); 176 if (ContentTypes.HasSource) { 177 OS << " source: "; 178 FileEntry.Source.dump(OS, DumpOptions); 179 OS << '\n'; 180 } 181 } 182 } 183 } 184 185 // Parse v2-v4 directory and file tables. 186 static Error 187 parseV2DirFileTables(const DWARFDataExtractor &DebugLineData, 188 uint64_t *OffsetPtr, 189 DWARFDebugLine::ContentTypeTracker &ContentTypes, 190 std::vector<DWARFFormValue> &IncludeDirectories, 191 std::vector<DWARFDebugLine::FileNameEntry> &FileNames) { 192 while (true) { 193 Error Err = Error::success(); 194 StringRef S = DebugLineData.getCStrRef(OffsetPtr, &Err); 195 if (Err) { 196 consumeError(std::move(Err)); 197 return createStringError(errc::invalid_argument, 198 "include directories table was not null " 199 "terminated before the end of the prologue"); 200 } 201 if (S.empty()) 202 break; 203 DWARFFormValue Dir = 204 DWARFFormValue::createFromPValue(dwarf::DW_FORM_string, S.data()); 205 IncludeDirectories.push_back(Dir); 206 } 207 208 ContentTypes.HasModTime = true; 209 ContentTypes.HasLength = true; 210 211 while (true) { 212 Error Err = Error::success(); 213 StringRef Name = DebugLineData.getCStrRef(OffsetPtr, &Err); 214 if (!Err && Name.empty()) 215 break; 216 217 DWARFDebugLine::FileNameEntry FileEntry; 218 FileEntry.Name = 219 DWARFFormValue::createFromPValue(dwarf::DW_FORM_string, Name.data()); 220 FileEntry.DirIdx = DebugLineData.getULEB128(OffsetPtr, &Err); 221 FileEntry.ModTime = DebugLineData.getULEB128(OffsetPtr, &Err); 222 FileEntry.Length = DebugLineData.getULEB128(OffsetPtr, &Err); 223 224 if (Err) { 225 consumeError(std::move(Err)); 226 return createStringError( 227 errc::invalid_argument, 228 "file names table was not null terminated before " 229 "the end of the prologue"); 230 } 231 FileNames.push_back(FileEntry); 232 } 233 234 return Error::success(); 235 } 236 237 // Parse v5 directory/file entry content descriptions. 238 // Returns the descriptors, or an error if we did not find a path or ran off 239 // the end of the prologue. 240 static llvm::Expected<ContentDescriptors> 241 parseV5EntryFormat(const DWARFDataExtractor &DebugLineData, uint64_t *OffsetPtr, 242 DWARFDebugLine::ContentTypeTracker *ContentTypes) { 243 Error Err = Error::success(); 244 ContentDescriptors Descriptors; 245 int FormatCount = DebugLineData.getU8(OffsetPtr, &Err); 246 bool HasPath = false; 247 for (int I = 0; I != FormatCount && !Err; ++I) { 248 ContentDescriptor Descriptor; 249 Descriptor.Type = 250 dwarf::LineNumberEntryFormat(DebugLineData.getULEB128(OffsetPtr, &Err)); 251 Descriptor.Form = dwarf::Form(DebugLineData.getULEB128(OffsetPtr, &Err)); 252 if (Descriptor.Type == dwarf::DW_LNCT_path) 253 HasPath = true; 254 if (ContentTypes) 255 ContentTypes->trackContentType(Descriptor.Type); 256 Descriptors.push_back(Descriptor); 257 } 258 259 if (Err) 260 return createStringError(errc::invalid_argument, 261 "failed to parse entry content descriptors: %s", 262 toString(std::move(Err)).c_str()); 263 264 if (!HasPath) 265 return createStringError(errc::invalid_argument, 266 "failed to parse entry content descriptions" 267 " because no path was found"); 268 return Descriptors; 269 } 270 271 static Error 272 parseV5DirFileTables(const DWARFDataExtractor &DebugLineData, 273 uint64_t *OffsetPtr, const dwarf::FormParams &FormParams, 274 const DWARFContext &Ctx, const DWARFUnit *U, 275 DWARFDebugLine::ContentTypeTracker &ContentTypes, 276 std::vector<DWARFFormValue> &IncludeDirectories, 277 std::vector<DWARFDebugLine::FileNameEntry> &FileNames) { 278 // Get the directory entry description. 279 llvm::Expected<ContentDescriptors> DirDescriptors = 280 parseV5EntryFormat(DebugLineData, OffsetPtr, nullptr); 281 if (!DirDescriptors) 282 return DirDescriptors.takeError(); 283 284 // Get the directory entries, according to the format described above. 285 uint64_t DirEntryCount = DebugLineData.getULEB128(OffsetPtr); 286 for (uint64_t I = 0; I != DirEntryCount; ++I) { 287 for (auto Descriptor : *DirDescriptors) { 288 DWARFFormValue Value(Descriptor.Form); 289 switch (Descriptor.Type) { 290 case DW_LNCT_path: 291 if (!Value.extractValue(DebugLineData, OffsetPtr, FormParams, &Ctx, U)) 292 return createStringError(errc::invalid_argument, 293 "failed to parse directory entry because " 294 "extracting the form value failed"); 295 IncludeDirectories.push_back(Value); 296 break; 297 default: 298 if (!Value.skipValue(DebugLineData, OffsetPtr, FormParams)) 299 return createStringError(errc::invalid_argument, 300 "failed to parse directory entry because " 301 "skipping the form value failed"); 302 } 303 } 304 } 305 306 // Get the file entry description. 307 llvm::Expected<ContentDescriptors> FileDescriptors = 308 parseV5EntryFormat(DebugLineData, OffsetPtr, &ContentTypes); 309 if (!FileDescriptors) 310 return FileDescriptors.takeError(); 311 312 // Get the file entries, according to the format described above. 313 uint64_t FileEntryCount = DebugLineData.getULEB128(OffsetPtr); 314 for (uint64_t I = 0; I != FileEntryCount; ++I) { 315 DWARFDebugLine::FileNameEntry FileEntry; 316 for (auto Descriptor : *FileDescriptors) { 317 DWARFFormValue Value(Descriptor.Form); 318 if (!Value.extractValue(DebugLineData, OffsetPtr, FormParams, &Ctx, U)) 319 return createStringError(errc::invalid_argument, 320 "failed to parse file entry because " 321 "extracting the form value failed"); 322 switch (Descriptor.Type) { 323 case DW_LNCT_path: 324 FileEntry.Name = Value; 325 break; 326 case DW_LNCT_LLVM_source: 327 FileEntry.Source = Value; 328 break; 329 case DW_LNCT_directory_index: 330 FileEntry.DirIdx = Value.getAsUnsignedConstant().getValue(); 331 break; 332 case DW_LNCT_timestamp: 333 FileEntry.ModTime = Value.getAsUnsignedConstant().getValue(); 334 break; 335 case DW_LNCT_size: 336 FileEntry.Length = Value.getAsUnsignedConstant().getValue(); 337 break; 338 case DW_LNCT_MD5: 339 if (!Value.getAsBlock() || Value.getAsBlock().getValue().size() != 16) 340 return createStringError( 341 errc::invalid_argument, 342 "failed to parse file entry because the MD5 hash is invalid"); 343 std::uninitialized_copy_n(Value.getAsBlock().getValue().begin(), 16, 344 FileEntry.Checksum.begin()); 345 break; 346 default: 347 break; 348 } 349 } 350 FileNames.push_back(FileEntry); 351 } 352 return Error::success(); 353 } 354 355 uint64_t DWARFDebugLine::Prologue::getLength() const { 356 uint64_t Length = PrologueLength + sizeofTotalLength() + 357 sizeof(getVersion()) + sizeofPrologueLength(); 358 if (getVersion() >= 5) 359 Length += 2; // Address + Segment selector sizes. 360 return Length; 361 } 362 363 Error DWARFDebugLine::Prologue::parse( 364 DWARFDataExtractor DebugLineData, uint64_t *OffsetPtr, 365 function_ref<void(Error)> RecoverableErrorHandler, const DWARFContext &Ctx, 366 const DWARFUnit *U) { 367 const uint64_t PrologueOffset = *OffsetPtr; 368 369 clear(); 370 DataExtractor::Cursor Cursor(*OffsetPtr); 371 std::tie(TotalLength, FormParams.Format) = 372 DebugLineData.getInitialLength(Cursor); 373 374 DebugLineData = 375 DWARFDataExtractor(DebugLineData, Cursor.tell() + TotalLength); 376 FormParams.Version = DebugLineData.getU16(Cursor); 377 if (Cursor && !versionIsSupported(getVersion())) { 378 // Treat this error as unrecoverable - we cannot be sure what any of 379 // the data represents including the length field, so cannot skip it or make 380 // any reasonable assumptions. 381 *OffsetPtr = Cursor.tell(); 382 return createStringError( 383 errc::not_supported, 384 "parsing line table prologue at offset 0x%8.8" PRIx64 385 ": unsupported version %" PRIu16, 386 PrologueOffset, getVersion()); 387 } 388 389 if (getVersion() >= 5) { 390 FormParams.AddrSize = DebugLineData.getU8(Cursor); 391 assert((!Cursor || DebugLineData.getAddressSize() == 0 || 392 DebugLineData.getAddressSize() == getAddressSize()) && 393 "Line table header and data extractor disagree"); 394 SegSelectorSize = DebugLineData.getU8(Cursor); 395 } 396 397 PrologueLength = 398 DebugLineData.getRelocatedValue(Cursor, sizeofPrologueLength()); 399 const uint64_t EndPrologueOffset = PrologueLength + Cursor.tell(); 400 DebugLineData = DWARFDataExtractor(DebugLineData, EndPrologueOffset); 401 MinInstLength = DebugLineData.getU8(Cursor); 402 if (getVersion() >= 4) 403 MaxOpsPerInst = DebugLineData.getU8(Cursor); 404 DefaultIsStmt = DebugLineData.getU8(Cursor); 405 LineBase = DebugLineData.getU8(Cursor); 406 LineRange = DebugLineData.getU8(Cursor); 407 OpcodeBase = DebugLineData.getU8(Cursor); 408 409 if (Cursor && OpcodeBase == 0) { 410 // If the opcode base is 0, we cannot read the standard opcode lengths (of 411 // which there are supposed to be one fewer than the opcode base). Assume 412 // there are no standard opcodes and continue parsing. 413 RecoverableErrorHandler(createStringError( 414 errc::invalid_argument, 415 "parsing line table prologue at offset 0x%8.8" PRIx64 416 " found opcode base of 0. Assuming no standard opcodes", 417 PrologueOffset)); 418 } else if (Cursor) { 419 StandardOpcodeLengths.reserve(OpcodeBase - 1); 420 for (uint32_t I = 1; I < OpcodeBase; ++I) { 421 uint8_t OpLen = DebugLineData.getU8(Cursor); 422 StandardOpcodeLengths.push_back(OpLen); 423 } 424 } 425 426 *OffsetPtr = Cursor.tell(); 427 // A corrupt file name or directory table does not prevent interpretation of 428 // the main line program, so check the cursor state now so that its errors can 429 // be handled separately. 430 if (!Cursor) 431 return createStringError( 432 errc::invalid_argument, 433 "parsing line table prologue at offset 0x%8.8" PRIx64 ": %s", 434 PrologueOffset, toString(Cursor.takeError()).c_str()); 435 436 Error E = 437 getVersion() >= 5 438 ? parseV5DirFileTables(DebugLineData, OffsetPtr, FormParams, Ctx, U, 439 ContentTypes, IncludeDirectories, FileNames) 440 : parseV2DirFileTables(DebugLineData, OffsetPtr, ContentTypes, 441 IncludeDirectories, FileNames); 442 if (E) { 443 RecoverableErrorHandler(joinErrors( 444 createStringError( 445 errc::invalid_argument, 446 "parsing line table prologue at 0x%8.8" PRIx64 447 " found an invalid directory or file table description at" 448 " 0x%8.8" PRIx64, 449 PrologueOffset, *OffsetPtr), 450 std::move(E))); 451 return Error::success(); 452 } 453 454 assert(*OffsetPtr <= EndPrologueOffset); 455 if (*OffsetPtr != EndPrologueOffset) { 456 RecoverableErrorHandler(createStringError( 457 errc::invalid_argument, 458 "unknown data in line table prologue at offset 0x%8.8" PRIx64 459 ": parsing ended (at offset 0x%8.8" PRIx64 460 ") before reaching the prologue end at offset 0x%8.8" PRIx64, 461 PrologueOffset, *OffsetPtr, EndPrologueOffset)); 462 } 463 return Error::success(); 464 } 465 466 DWARFDebugLine::Row::Row(bool DefaultIsStmt) { reset(DefaultIsStmt); } 467 468 void DWARFDebugLine::Row::postAppend() { 469 Discriminator = 0; 470 BasicBlock = false; 471 PrologueEnd = false; 472 EpilogueBegin = false; 473 } 474 475 void DWARFDebugLine::Row::reset(bool DefaultIsStmt) { 476 Address.Address = 0; 477 Address.SectionIndex = object::SectionedAddress::UndefSection; 478 Line = 1; 479 Column = 0; 480 File = 1; 481 Isa = 0; 482 Discriminator = 0; 483 IsStmt = DefaultIsStmt; 484 BasicBlock = false; 485 EndSequence = false; 486 PrologueEnd = false; 487 EpilogueBegin = false; 488 } 489 490 void DWARFDebugLine::Row::dumpTableHeader(raw_ostream &OS, unsigned Indent) { 491 OS.indent(Indent) 492 << "Address Line Column File ISA Discriminator Flags\n"; 493 OS.indent(Indent) 494 << "------------------ ------ ------ ------ --- ------------- " 495 "-------------\n"; 496 } 497 498 void DWARFDebugLine::Row::dump(raw_ostream &OS) const { 499 OS << format("0x%16.16" PRIx64 " %6u %6u", Address.Address, Line, Column) 500 << format(" %6u %3u %13u ", File, Isa, Discriminator) 501 << (IsStmt ? " is_stmt" : "") << (BasicBlock ? " basic_block" : "") 502 << (PrologueEnd ? " prologue_end" : "") 503 << (EpilogueBegin ? " epilogue_begin" : "") 504 << (EndSequence ? " end_sequence" : "") << '\n'; 505 } 506 507 DWARFDebugLine::Sequence::Sequence() { reset(); } 508 509 void DWARFDebugLine::Sequence::reset() { 510 LowPC = 0; 511 HighPC = 0; 512 SectionIndex = object::SectionedAddress::UndefSection; 513 FirstRowIndex = 0; 514 LastRowIndex = 0; 515 Empty = true; 516 } 517 518 DWARFDebugLine::LineTable::LineTable() { clear(); } 519 520 void DWARFDebugLine::LineTable::dump(raw_ostream &OS, 521 DIDumpOptions DumpOptions) const { 522 Prologue.dump(OS, DumpOptions); 523 524 if (!Rows.empty()) { 525 OS << '\n'; 526 Row::dumpTableHeader(OS, 0); 527 for (const Row &R : Rows) { 528 R.dump(OS); 529 } 530 } 531 532 // Terminate the table with a final blank line to clearly delineate it from 533 // later dumps. 534 OS << '\n'; 535 } 536 537 void DWARFDebugLine::LineTable::clear() { 538 Prologue.clear(); 539 Rows.clear(); 540 Sequences.clear(); 541 } 542 543 DWARFDebugLine::ParsingState::ParsingState( 544 struct LineTable *LT, uint64_t TableOffset, 545 function_ref<void(Error)> ErrorHandler) 546 : LineTable(LT), LineTableOffset(TableOffset), ErrorHandler(ErrorHandler) { 547 resetRowAndSequence(); 548 } 549 550 void DWARFDebugLine::ParsingState::resetRowAndSequence() { 551 Row.reset(LineTable->Prologue.DefaultIsStmt); 552 Sequence.reset(); 553 } 554 555 void DWARFDebugLine::ParsingState::appendRowToMatrix() { 556 unsigned RowNumber = LineTable->Rows.size(); 557 if (Sequence.Empty) { 558 // Record the beginning of instruction sequence. 559 Sequence.Empty = false; 560 Sequence.LowPC = Row.Address.Address; 561 Sequence.FirstRowIndex = RowNumber; 562 } 563 LineTable->appendRow(Row); 564 if (Row.EndSequence) { 565 // Record the end of instruction sequence. 566 Sequence.HighPC = Row.Address.Address; 567 Sequence.LastRowIndex = RowNumber + 1; 568 Sequence.SectionIndex = Row.Address.SectionIndex; 569 if (Sequence.isValid()) 570 LineTable->appendSequence(Sequence); 571 Sequence.reset(); 572 } 573 Row.postAppend(); 574 } 575 576 const DWARFDebugLine::LineTable * 577 DWARFDebugLine::getLineTable(uint64_t Offset) const { 578 LineTableConstIter Pos = LineTableMap.find(Offset); 579 if (Pos != LineTableMap.end()) 580 return &Pos->second; 581 return nullptr; 582 } 583 584 Expected<const DWARFDebugLine::LineTable *> DWARFDebugLine::getOrParseLineTable( 585 DWARFDataExtractor &DebugLineData, uint64_t Offset, const DWARFContext &Ctx, 586 const DWARFUnit *U, function_ref<void(Error)> RecoverableErrorHandler) { 587 if (!DebugLineData.isValidOffset(Offset)) 588 return createStringError(errc::invalid_argument, "offset 0x%8.8" PRIx64 589 " is not a valid debug line section offset", 590 Offset); 591 592 std::pair<LineTableIter, bool> Pos = 593 LineTableMap.insert(LineTableMapTy::value_type(Offset, LineTable())); 594 LineTable *LT = &Pos.first->second; 595 if (Pos.second) { 596 if (Error Err = 597 LT->parse(DebugLineData, &Offset, Ctx, U, RecoverableErrorHandler)) 598 return std::move(Err); 599 return LT; 600 } 601 return LT; 602 } 603 604 static StringRef getOpcodeName(uint8_t Opcode, uint8_t OpcodeBase) { 605 assert(Opcode != 0); 606 if (Opcode < OpcodeBase) 607 return LNStandardString(Opcode); 608 return "special"; 609 } 610 611 uint64_t DWARFDebugLine::ParsingState::advanceAddr(uint64_t OperationAdvance, 612 uint8_t Opcode, 613 uint64_t OpcodeOffset) { 614 StringRef OpcodeName = getOpcodeName(Opcode, LineTable->Prologue.OpcodeBase); 615 // For versions less than 4, the MaxOpsPerInst member is set to 0, as the 616 // maximum_operations_per_instruction field wasn't introduced until DWARFv4. 617 // Don't warn about bad values in this situation. 618 if (ReportAdvanceAddrProblem && LineTable->Prologue.getVersion() >= 4 && 619 LineTable->Prologue.MaxOpsPerInst != 1) 620 ErrorHandler(createStringError( 621 errc::not_supported, 622 "line table program at offset 0x%8.8" PRIx64 623 " contains a %s opcode at offset 0x%8.8" PRIx64 624 ", but the prologue maximum_operations_per_instruction value is %" PRId8 625 ", which is unsupported. Assuming a value of 1 instead", 626 LineTableOffset, OpcodeName.data(), OpcodeOffset, 627 LineTable->Prologue.MaxOpsPerInst)); 628 if (ReportAdvanceAddrProblem && LineTable->Prologue.MinInstLength == 0) 629 ErrorHandler( 630 createStringError(errc::invalid_argument, 631 "line table program at offset 0x%8.8" PRIx64 632 " contains a %s opcode at offset 0x%8.8" PRIx64 633 ", but the prologue minimum_instruction_length value " 634 "is 0, which prevents any address advancing", 635 LineTableOffset, OpcodeName.data(), OpcodeOffset)); 636 ReportAdvanceAddrProblem = false; 637 uint64_t AddrOffset = OperationAdvance * LineTable->Prologue.MinInstLength; 638 Row.Address.Address += AddrOffset; 639 return AddrOffset; 640 } 641 642 DWARFDebugLine::ParsingState::AddrAndAdjustedOpcode 643 DWARFDebugLine::ParsingState::advanceAddrForOpcode(uint8_t Opcode, 644 uint64_t OpcodeOffset) { 645 assert(Opcode == DW_LNS_const_add_pc || 646 Opcode >= LineTable->Prologue.OpcodeBase); 647 if (ReportBadLineRange && LineTable->Prologue.LineRange == 0) { 648 StringRef OpcodeName = 649 getOpcodeName(Opcode, LineTable->Prologue.OpcodeBase); 650 ErrorHandler( 651 createStringError(errc::not_supported, 652 "line table program at offset 0x%8.8" PRIx64 653 " contains a %s opcode at offset 0x%8.8" PRIx64 654 ", but the prologue line_range value is 0. The " 655 "address and line will not be adjusted", 656 LineTableOffset, OpcodeName.data(), OpcodeOffset)); 657 ReportBadLineRange = false; 658 } 659 660 uint8_t OpcodeValue = Opcode; 661 if (Opcode == DW_LNS_const_add_pc) 662 OpcodeValue = 255; 663 uint8_t AdjustedOpcode = OpcodeValue - LineTable->Prologue.OpcodeBase; 664 uint64_t OperationAdvance = 665 LineTable->Prologue.LineRange != 0 666 ? AdjustedOpcode / LineTable->Prologue.LineRange 667 : 0; 668 uint64_t AddrOffset = advanceAddr(OperationAdvance, Opcode, OpcodeOffset); 669 return {AddrOffset, AdjustedOpcode}; 670 } 671 672 DWARFDebugLine::ParsingState::AddrAndLineDelta 673 DWARFDebugLine::ParsingState::handleSpecialOpcode(uint8_t Opcode, 674 uint64_t OpcodeOffset) { 675 // A special opcode value is chosen based on the amount that needs 676 // to be added to the line and address registers. The maximum line 677 // increment for a special opcode is the value of the line_base 678 // field in the header, plus the value of the line_range field, 679 // minus 1 (line base + line range - 1). If the desired line 680 // increment is greater than the maximum line increment, a standard 681 // opcode must be used instead of a special opcode. The "address 682 // advance" is calculated by dividing the desired address increment 683 // by the minimum_instruction_length field from the header. The 684 // special opcode is then calculated using the following formula: 685 // 686 // opcode = (desired line increment - line_base) + 687 // (line_range * address advance) + opcode_base 688 // 689 // If the resulting opcode is greater than 255, a standard opcode 690 // must be used instead. 691 // 692 // To decode a special opcode, subtract the opcode_base from the 693 // opcode itself to give the adjusted opcode. The amount to 694 // increment the address register is the result of the adjusted 695 // opcode divided by the line_range multiplied by the 696 // minimum_instruction_length field from the header. That is: 697 // 698 // address increment = (adjusted opcode / line_range) * 699 // minimum_instruction_length 700 // 701 // The amount to increment the line register is the line_base plus 702 // the result of the adjusted opcode modulo the line_range. That is: 703 // 704 // line increment = line_base + (adjusted opcode % line_range) 705 706 DWARFDebugLine::ParsingState::AddrAndAdjustedOpcode AddrAdvanceResult = 707 advanceAddrForOpcode(Opcode, OpcodeOffset); 708 int32_t LineOffset = 0; 709 if (LineTable->Prologue.LineRange != 0) 710 LineOffset = 711 LineTable->Prologue.LineBase + 712 (AddrAdvanceResult.AdjustedOpcode % LineTable->Prologue.LineRange); 713 Row.Line += LineOffset; 714 return {AddrAdvanceResult.AddrDelta, LineOffset}; 715 } 716 717 /// Parse a ULEB128 using the specified \p Cursor. \returns the parsed value on 718 /// success, or None if \p Cursor is in a failing state. 719 template <typename T> 720 static Optional<T> parseULEB128(DWARFDataExtractor &Data, 721 DataExtractor::Cursor &Cursor) { 722 T Value = Data.getULEB128(Cursor); 723 if (Cursor) 724 return Value; 725 return None; 726 } 727 728 Error DWARFDebugLine::LineTable::parse( 729 DWARFDataExtractor &DebugLineData, uint64_t *OffsetPtr, 730 const DWARFContext &Ctx, const DWARFUnit *U, 731 function_ref<void(Error)> RecoverableErrorHandler, raw_ostream *OS, 732 bool Verbose) { 733 assert((OS || !Verbose) && "cannot have verbose output without stream"); 734 const uint64_t DebugLineOffset = *OffsetPtr; 735 736 clear(); 737 738 Error PrologueErr = 739 Prologue.parse(DebugLineData, OffsetPtr, RecoverableErrorHandler, Ctx, U); 740 741 if (OS) { 742 DIDumpOptions DumpOptions; 743 DumpOptions.Verbose = Verbose; 744 Prologue.dump(*OS, DumpOptions); 745 } 746 747 if (PrologueErr) { 748 // Ensure there is a blank line after the prologue to clearly delineate it 749 // from later dumps. 750 if (OS) 751 *OS << "\n"; 752 return PrologueErr; 753 } 754 755 uint64_t ProgramLength = Prologue.TotalLength + Prologue.sizeofTotalLength(); 756 if (!DebugLineData.isValidOffsetForDataOfSize(DebugLineOffset, 757 ProgramLength)) { 758 assert(DebugLineData.size() > DebugLineOffset && 759 "prologue parsing should handle invalid offset"); 760 uint64_t BytesRemaining = DebugLineData.size() - DebugLineOffset; 761 RecoverableErrorHandler( 762 createStringError(errc::invalid_argument, 763 "line table program with offset 0x%8.8" PRIx64 764 " has length 0x%8.8" PRIx64 " but only 0x%8.8" PRIx64 765 " bytes are available", 766 DebugLineOffset, ProgramLength, BytesRemaining)); 767 // Continue by capping the length at the number of remaining bytes. 768 ProgramLength = BytesRemaining; 769 } 770 771 // Create a DataExtractor which can only see the data up to the end of the 772 // table, to prevent reading past the end. 773 const uint64_t EndOffset = DebugLineOffset + ProgramLength; 774 DWARFDataExtractor TableData(DebugLineData, EndOffset); 775 776 // See if we should tell the data extractor the address size. 777 if (TableData.getAddressSize() == 0) 778 TableData.setAddressSize(Prologue.getAddressSize()); 779 else 780 assert(Prologue.getAddressSize() == 0 || 781 Prologue.getAddressSize() == TableData.getAddressSize()); 782 783 ParsingState State(this, DebugLineOffset, RecoverableErrorHandler); 784 785 *OffsetPtr = DebugLineOffset + Prologue.getLength(); 786 if (OS && *OffsetPtr < EndOffset) { 787 *OS << '\n'; 788 Row::dumpTableHeader(*OS, /*Indent=*/Verbose ? 12 : 0); 789 } 790 bool TombstonedAddress = false; 791 auto EmitRow = [&] { 792 if (!TombstonedAddress) { 793 if (Verbose) { 794 *OS << "\n"; 795 OS->indent(12); 796 } 797 if (OS) 798 State.Row.dump(*OS); 799 State.appendRowToMatrix(); 800 } 801 }; 802 while (*OffsetPtr < EndOffset) { 803 DataExtractor::Cursor Cursor(*OffsetPtr); 804 805 if (Verbose) 806 *OS << format("0x%08.08" PRIx64 ": ", *OffsetPtr); 807 808 uint64_t OpcodeOffset = *OffsetPtr; 809 uint8_t Opcode = TableData.getU8(Cursor); 810 size_t RowCount = Rows.size(); 811 812 if (Cursor && Verbose) 813 *OS << format("%02.02" PRIx8 " ", Opcode); 814 815 if (Opcode == 0) { 816 // Extended Opcodes always start with a zero opcode followed by 817 // a uleb128 length so you can skip ones you don't know about 818 uint64_t Len = TableData.getULEB128(Cursor); 819 uint64_t ExtOffset = Cursor.tell(); 820 821 // Tolerate zero-length; assume length is correct and soldier on. 822 if (Len == 0) { 823 if (Cursor && Verbose) 824 *OS << "Badly formed extended line op (length 0)\n"; 825 if (!Cursor) { 826 if (Verbose) 827 *OS << "\n"; 828 RecoverableErrorHandler(Cursor.takeError()); 829 } 830 *OffsetPtr = Cursor.tell(); 831 continue; 832 } 833 834 uint8_t SubOpcode = TableData.getU8(Cursor); 835 // OperandOffset will be the same as ExtOffset, if it was not possible to 836 // read the SubOpcode. 837 uint64_t OperandOffset = Cursor.tell(); 838 if (Verbose) 839 *OS << LNExtendedString(SubOpcode); 840 switch (SubOpcode) { 841 case DW_LNE_end_sequence: 842 // Set the end_sequence register of the state machine to true and 843 // append a row to the matrix using the current values of the 844 // state-machine registers. Then reset the registers to the initial 845 // values specified above. Every statement program sequence must end 846 // with a DW_LNE_end_sequence instruction which creates a row whose 847 // address is that of the byte after the last target machine instruction 848 // of the sequence. 849 State.Row.EndSequence = true; 850 // No need to test the Cursor is valid here, since it must be to get 851 // into this code path - if it were invalid, the default case would be 852 // followed. 853 EmitRow(); 854 State.resetRowAndSequence(); 855 break; 856 857 case DW_LNE_set_address: 858 // Takes a single relocatable address as an operand. The size of the 859 // operand is the size appropriate to hold an address on the target 860 // machine. Set the address register to the value given by the 861 // relocatable address. All of the other statement program opcodes 862 // that affect the address register add a delta to it. This instruction 863 // stores a relocatable value into it instead. 864 // 865 // Make sure the extractor knows the address size. If not, infer it 866 // from the size of the operand. 867 { 868 uint8_t ExtractorAddressSize = TableData.getAddressSize(); 869 uint64_t OpcodeAddressSize = Len - 1; 870 if (ExtractorAddressSize != OpcodeAddressSize && 871 ExtractorAddressSize != 0) 872 RecoverableErrorHandler(createStringError( 873 errc::invalid_argument, 874 "mismatching address size at offset 0x%8.8" PRIx64 875 " expected 0x%2.2" PRIx8 " found 0x%2.2" PRIx64, 876 ExtOffset, ExtractorAddressSize, Len - 1)); 877 878 // Assume that the line table is correct and temporarily override the 879 // address size. If the size is unsupported, give up trying to read 880 // the address and continue to the next opcode. 881 if (OpcodeAddressSize != 1 && OpcodeAddressSize != 2 && 882 OpcodeAddressSize != 4 && OpcodeAddressSize != 8) { 883 RecoverableErrorHandler(createStringError( 884 errc::invalid_argument, 885 "address size 0x%2.2" PRIx64 886 " of DW_LNE_set_address opcode at offset 0x%8.8" PRIx64 887 " is unsupported", 888 OpcodeAddressSize, ExtOffset)); 889 TableData.skip(Cursor, OpcodeAddressSize); 890 } else { 891 TableData.setAddressSize(OpcodeAddressSize); 892 State.Row.Address.Address = TableData.getRelocatedAddress( 893 Cursor, &State.Row.Address.SectionIndex); 894 895 uint64_t Tombstone = 896 dwarf::computeTombstoneAddress(OpcodeAddressSize); 897 TombstonedAddress = State.Row.Address.Address == Tombstone; 898 899 // Restore the address size if the extractor already had it. 900 if (ExtractorAddressSize != 0) 901 TableData.setAddressSize(ExtractorAddressSize); 902 } 903 904 if (Cursor && Verbose) { 905 *OS << " ("; 906 DWARFFormValue::dumpAddress(*OS, OpcodeAddressSize, State.Row.Address.Address); 907 *OS << ')'; 908 } 909 } 910 break; 911 912 case DW_LNE_define_file: 913 // Takes 4 arguments. The first is a null terminated string containing 914 // a source file name. The second is an unsigned LEB128 number 915 // representing the directory index of the directory in which the file 916 // was found. The third is an unsigned LEB128 number representing the 917 // time of last modification of the file. The fourth is an unsigned 918 // LEB128 number representing the length in bytes of the file. The time 919 // and length fields may contain LEB128(0) if the information is not 920 // available. 921 // 922 // The directory index represents an entry in the include_directories 923 // section of the statement program prologue. The index is LEB128(0) 924 // if the file was found in the current directory of the compilation, 925 // LEB128(1) if it was found in the first directory in the 926 // include_directories section, and so on. The directory index is 927 // ignored for file names that represent full path names. 928 // 929 // The files are numbered, starting at 1, in the order in which they 930 // appear; the names in the prologue come before names defined by 931 // the DW_LNE_define_file instruction. These numbers are used in the 932 // the file register of the state machine. 933 { 934 FileNameEntry FileEntry; 935 const char *Name = TableData.getCStr(Cursor); 936 FileEntry.Name = 937 DWARFFormValue::createFromPValue(dwarf::DW_FORM_string, Name); 938 FileEntry.DirIdx = TableData.getULEB128(Cursor); 939 FileEntry.ModTime = TableData.getULEB128(Cursor); 940 FileEntry.Length = TableData.getULEB128(Cursor); 941 Prologue.FileNames.push_back(FileEntry); 942 if (Cursor && Verbose) 943 *OS << " (" << Name << ", dir=" << FileEntry.DirIdx << ", mod_time=" 944 << format("(0x%16.16" PRIx64 ")", FileEntry.ModTime) 945 << ", length=" << FileEntry.Length << ")"; 946 } 947 break; 948 949 case DW_LNE_set_discriminator: 950 State.Row.Discriminator = TableData.getULEB128(Cursor); 951 if (Cursor && Verbose) 952 *OS << " (" << State.Row.Discriminator << ")"; 953 break; 954 955 default: 956 if (Cursor && Verbose) 957 *OS << format("Unrecognized extended op 0x%02.02" PRIx8, SubOpcode) 958 << format(" length %" PRIx64, Len); 959 // Len doesn't include the zero opcode byte or the length itself, but 960 // it does include the sub_opcode, so we have to adjust for that. 961 TableData.skip(Cursor, Len - 1); 962 break; 963 } 964 // Make sure the length as recorded in the table and the standard length 965 // for the opcode match. If they don't, continue from the end as claimed 966 // by the table. Similarly, continue from the claimed end in the event of 967 // a parsing error. 968 uint64_t End = ExtOffset + Len; 969 if (Cursor && Cursor.tell() != End) 970 RecoverableErrorHandler(createStringError( 971 errc::illegal_byte_sequence, 972 "unexpected line op length at offset 0x%8.8" PRIx64 973 " expected 0x%2.2" PRIx64 " found 0x%2.2" PRIx64, 974 ExtOffset, Len, Cursor.tell() - ExtOffset)); 975 if (!Cursor && Verbose) { 976 DWARFDataExtractor::Cursor ByteCursor(OperandOffset); 977 uint8_t Byte = TableData.getU8(ByteCursor); 978 if (ByteCursor) { 979 *OS << " (<parsing error>"; 980 do { 981 *OS << format(" %2.2" PRIx8, Byte); 982 Byte = TableData.getU8(ByteCursor); 983 } while (ByteCursor); 984 *OS << ")"; 985 } 986 987 // The only parse failure in this case should be if the end was reached. 988 // In that case, throw away the error, as the main Cursor's error will 989 // be sufficient. 990 consumeError(ByteCursor.takeError()); 991 } 992 *OffsetPtr = End; 993 } else if (Opcode < Prologue.OpcodeBase) { 994 if (Verbose) 995 *OS << LNStandardString(Opcode); 996 switch (Opcode) { 997 // Standard Opcodes 998 case DW_LNS_copy: 999 // Takes no arguments. Append a row to the matrix using the 1000 // current values of the state-machine registers. 1001 EmitRow(); 1002 break; 1003 1004 case DW_LNS_advance_pc: 1005 // Takes a single unsigned LEB128 operand, multiplies it by the 1006 // min_inst_length field of the prologue, and adds the 1007 // result to the address register of the state machine. 1008 if (Optional<uint64_t> Operand = 1009 parseULEB128<uint64_t>(TableData, Cursor)) { 1010 uint64_t AddrOffset = 1011 State.advanceAddr(*Operand, Opcode, OpcodeOffset); 1012 if (Verbose) 1013 *OS << " (" << AddrOffset << ")"; 1014 } 1015 break; 1016 1017 case DW_LNS_advance_line: 1018 // Takes a single signed LEB128 operand and adds that value to 1019 // the line register of the state machine. 1020 { 1021 int64_t LineDelta = TableData.getSLEB128(Cursor); 1022 if (Cursor) { 1023 State.Row.Line += LineDelta; 1024 if (Verbose) 1025 *OS << " (" << State.Row.Line << ")"; 1026 } 1027 } 1028 break; 1029 1030 case DW_LNS_set_file: 1031 // Takes a single unsigned LEB128 operand and stores it in the file 1032 // register of the state machine. 1033 if (Optional<uint16_t> File = 1034 parseULEB128<uint16_t>(TableData, Cursor)) { 1035 State.Row.File = *File; 1036 if (Verbose) 1037 *OS << " (" << State.Row.File << ")"; 1038 } 1039 break; 1040 1041 case DW_LNS_set_column: 1042 // Takes a single unsigned LEB128 operand and stores it in the 1043 // column register of the state machine. 1044 if (Optional<uint16_t> Column = 1045 parseULEB128<uint16_t>(TableData, Cursor)) { 1046 State.Row.Column = *Column; 1047 if (Verbose) 1048 *OS << " (" << State.Row.Column << ")"; 1049 } 1050 break; 1051 1052 case DW_LNS_negate_stmt: 1053 // Takes no arguments. Set the is_stmt register of the state 1054 // machine to the logical negation of its current value. 1055 State.Row.IsStmt = !State.Row.IsStmt; 1056 break; 1057 1058 case DW_LNS_set_basic_block: 1059 // Takes no arguments. Set the basic_block register of the 1060 // state machine to true 1061 State.Row.BasicBlock = true; 1062 break; 1063 1064 case DW_LNS_const_add_pc: 1065 // Takes no arguments. Add to the address register of the state 1066 // machine the address increment value corresponding to special 1067 // opcode 255. The motivation for DW_LNS_const_add_pc is this: 1068 // when the statement program needs to advance the address by a 1069 // small amount, it can use a single special opcode, which occupies 1070 // a single byte. When it needs to advance the address by up to 1071 // twice the range of the last special opcode, it can use 1072 // DW_LNS_const_add_pc followed by a special opcode, for a total 1073 // of two bytes. Only if it needs to advance the address by more 1074 // than twice that range will it need to use both DW_LNS_advance_pc 1075 // and a special opcode, requiring three or more bytes. 1076 { 1077 uint64_t AddrOffset = 1078 State.advanceAddrForOpcode(Opcode, OpcodeOffset).AddrDelta; 1079 if (Verbose) 1080 *OS << format(" (0x%16.16" PRIx64 ")", AddrOffset); 1081 } 1082 break; 1083 1084 case DW_LNS_fixed_advance_pc: 1085 // Takes a single uhalf operand. Add to the address register of 1086 // the state machine the value of the (unencoded) operand. This 1087 // is the only extended opcode that takes an argument that is not 1088 // a variable length number. The motivation for DW_LNS_fixed_advance_pc 1089 // is this: existing assemblers cannot emit DW_LNS_advance_pc or 1090 // special opcodes because they cannot encode LEB128 numbers or 1091 // judge when the computation of a special opcode overflows and 1092 // requires the use of DW_LNS_advance_pc. Such assemblers, however, 1093 // can use DW_LNS_fixed_advance_pc instead, sacrificing compression. 1094 { 1095 uint16_t PCOffset = 1096 TableData.getRelocatedValue(Cursor, 2); 1097 if (Cursor) { 1098 State.Row.Address.Address += PCOffset; 1099 if (Verbose) 1100 *OS << format(" (0x%4.4" PRIx16 ")", PCOffset); 1101 } 1102 } 1103 break; 1104 1105 case DW_LNS_set_prologue_end: 1106 // Takes no arguments. Set the prologue_end register of the 1107 // state machine to true 1108 State.Row.PrologueEnd = true; 1109 break; 1110 1111 case DW_LNS_set_epilogue_begin: 1112 // Takes no arguments. Set the basic_block register of the 1113 // state machine to true 1114 State.Row.EpilogueBegin = true; 1115 break; 1116 1117 case DW_LNS_set_isa: 1118 // Takes a single unsigned LEB128 operand and stores it in the 1119 // ISA register of the state machine. 1120 if (Optional<uint8_t> Isa = parseULEB128<uint8_t>(TableData, Cursor)) { 1121 State.Row.Isa = *Isa; 1122 if (Verbose) 1123 *OS << " (" << (uint64_t)State.Row.Isa << ")"; 1124 } 1125 break; 1126 1127 default: 1128 // Handle any unknown standard opcodes here. We know the lengths 1129 // of such opcodes because they are specified in the prologue 1130 // as a multiple of LEB128 operands for each opcode. 1131 { 1132 assert(Opcode - 1U < Prologue.StandardOpcodeLengths.size()); 1133 if (Verbose) 1134 *OS << "Unrecognized standard opcode"; 1135 uint8_t OpcodeLength = Prologue.StandardOpcodeLengths[Opcode - 1]; 1136 std::vector<uint64_t> Operands; 1137 for (uint8_t I = 0; I < OpcodeLength; ++I) { 1138 if (Optional<uint64_t> Value = 1139 parseULEB128<uint64_t>(TableData, Cursor)) 1140 Operands.push_back(*Value); 1141 else 1142 break; 1143 } 1144 if (Verbose && !Operands.empty()) { 1145 *OS << " (operands: "; 1146 bool First = true; 1147 for (uint64_t Value : Operands) { 1148 if (!First) 1149 *OS << ", "; 1150 First = false; 1151 *OS << format("0x%16.16" PRIx64, Value); 1152 } 1153 if (Verbose) 1154 *OS << ')'; 1155 } 1156 } 1157 break; 1158 } 1159 1160 *OffsetPtr = Cursor.tell(); 1161 } else { 1162 // Special Opcodes. 1163 ParsingState::AddrAndLineDelta Delta = 1164 State.handleSpecialOpcode(Opcode, OpcodeOffset); 1165 1166 if (Verbose) 1167 *OS << "address += " << Delta.Address << ", line += " << Delta.Line; 1168 EmitRow(); 1169 *OffsetPtr = Cursor.tell(); 1170 } 1171 1172 // When a row is added to the matrix, it is also dumped, which includes a 1173 // new line already, so don't add an extra one. 1174 if (Verbose && Rows.size() == RowCount) 1175 *OS << "\n"; 1176 1177 // Most parse failures other than when parsing extended opcodes are due to 1178 // failures to read ULEBs. Bail out of parsing, since we don't know where to 1179 // continue reading from as there is no stated length for such byte 1180 // sequences. Print the final trailing new line if needed before doing so. 1181 if (!Cursor && Opcode != 0) { 1182 if (Verbose) 1183 *OS << "\n"; 1184 return Cursor.takeError(); 1185 } 1186 1187 if (!Cursor) 1188 RecoverableErrorHandler(Cursor.takeError()); 1189 } 1190 1191 if (!State.Sequence.Empty) 1192 RecoverableErrorHandler(createStringError( 1193 errc::illegal_byte_sequence, 1194 "last sequence in debug line table at offset 0x%8.8" PRIx64 1195 " is not terminated", 1196 DebugLineOffset)); 1197 1198 // Sort all sequences so that address lookup will work faster. 1199 if (!Sequences.empty()) { 1200 llvm::sort(Sequences, Sequence::orderByHighPC); 1201 // Note: actually, instruction address ranges of sequences should not 1202 // overlap (in shared objects and executables). If they do, the address 1203 // lookup would still work, though, but result would be ambiguous. 1204 // We don't report warning in this case. For example, 1205 // sometimes .so compiled from multiple object files contains a few 1206 // rudimentary sequences for address ranges [0x0, 0xsomething). 1207 } 1208 1209 // Terminate the table with a final blank line to clearly delineate it from 1210 // later dumps. 1211 if (OS) 1212 *OS << "\n"; 1213 1214 return Error::success(); 1215 } 1216 1217 uint32_t DWARFDebugLine::LineTable::findRowInSeq( 1218 const DWARFDebugLine::Sequence &Seq, 1219 object::SectionedAddress Address) const { 1220 if (!Seq.containsPC(Address)) 1221 return UnknownRowIndex; 1222 assert(Seq.SectionIndex == Address.SectionIndex); 1223 // In some cases, e.g. first instruction in a function, the compiler generates 1224 // two entries, both with the same address. We want the last one. 1225 // 1226 // In general we want a non-empty range: the last row whose address is less 1227 // than or equal to Address. This can be computed as upper_bound - 1. 1228 DWARFDebugLine::Row Row; 1229 Row.Address = Address; 1230 RowIter FirstRow = Rows.begin() + Seq.FirstRowIndex; 1231 RowIter LastRow = Rows.begin() + Seq.LastRowIndex; 1232 assert(FirstRow->Address.Address <= Row.Address.Address && 1233 Row.Address.Address < LastRow[-1].Address.Address); 1234 RowIter RowPos = std::upper_bound(FirstRow + 1, LastRow - 1, Row, 1235 DWARFDebugLine::Row::orderByAddress) - 1236 1; 1237 assert(Seq.SectionIndex == RowPos->Address.SectionIndex); 1238 return RowPos - Rows.begin(); 1239 } 1240 1241 uint32_t DWARFDebugLine::LineTable::lookupAddress( 1242 object::SectionedAddress Address) const { 1243 1244 // Search for relocatable addresses 1245 uint32_t Result = lookupAddressImpl(Address); 1246 1247 if (Result != UnknownRowIndex || 1248 Address.SectionIndex == object::SectionedAddress::UndefSection) 1249 return Result; 1250 1251 // Search for absolute addresses 1252 Address.SectionIndex = object::SectionedAddress::UndefSection; 1253 return lookupAddressImpl(Address); 1254 } 1255 1256 uint32_t DWARFDebugLine::LineTable::lookupAddressImpl( 1257 object::SectionedAddress Address) const { 1258 // First, find an instruction sequence containing the given address. 1259 DWARFDebugLine::Sequence Sequence; 1260 Sequence.SectionIndex = Address.SectionIndex; 1261 Sequence.HighPC = Address.Address; 1262 SequenceIter It = llvm::upper_bound(Sequences, Sequence, 1263 DWARFDebugLine::Sequence::orderByHighPC); 1264 if (It == Sequences.end() || It->SectionIndex != Address.SectionIndex) 1265 return UnknownRowIndex; 1266 return findRowInSeq(*It, Address); 1267 } 1268 1269 bool DWARFDebugLine::LineTable::lookupAddressRange( 1270 object::SectionedAddress Address, uint64_t Size, 1271 std::vector<uint32_t> &Result) const { 1272 1273 // Search for relocatable addresses 1274 if (lookupAddressRangeImpl(Address, Size, Result)) 1275 return true; 1276 1277 if (Address.SectionIndex == object::SectionedAddress::UndefSection) 1278 return false; 1279 1280 // Search for absolute addresses 1281 Address.SectionIndex = object::SectionedAddress::UndefSection; 1282 return lookupAddressRangeImpl(Address, Size, Result); 1283 } 1284 1285 bool DWARFDebugLine::LineTable::lookupAddressRangeImpl( 1286 object::SectionedAddress Address, uint64_t Size, 1287 std::vector<uint32_t> &Result) const { 1288 if (Sequences.empty()) 1289 return false; 1290 uint64_t EndAddr = Address.Address + Size; 1291 // First, find an instruction sequence containing the given address. 1292 DWARFDebugLine::Sequence Sequence; 1293 Sequence.SectionIndex = Address.SectionIndex; 1294 Sequence.HighPC = Address.Address; 1295 SequenceIter LastSeq = Sequences.end(); 1296 SequenceIter SeqPos = llvm::upper_bound( 1297 Sequences, Sequence, DWARFDebugLine::Sequence::orderByHighPC); 1298 if (SeqPos == LastSeq || !SeqPos->containsPC(Address)) 1299 return false; 1300 1301 SequenceIter StartPos = SeqPos; 1302 1303 // Add the rows from the first sequence to the vector, starting with the 1304 // index we just calculated 1305 1306 while (SeqPos != LastSeq && SeqPos->LowPC < EndAddr) { 1307 const DWARFDebugLine::Sequence &CurSeq = *SeqPos; 1308 // For the first sequence, we need to find which row in the sequence is the 1309 // first in our range. 1310 uint32_t FirstRowIndex = CurSeq.FirstRowIndex; 1311 if (SeqPos == StartPos) 1312 FirstRowIndex = findRowInSeq(CurSeq, Address); 1313 1314 // Figure out the last row in the range. 1315 uint32_t LastRowIndex = 1316 findRowInSeq(CurSeq, {EndAddr - 1, Address.SectionIndex}); 1317 if (LastRowIndex == UnknownRowIndex) 1318 LastRowIndex = CurSeq.LastRowIndex - 1; 1319 1320 assert(FirstRowIndex != UnknownRowIndex); 1321 assert(LastRowIndex != UnknownRowIndex); 1322 1323 for (uint32_t I = FirstRowIndex; I <= LastRowIndex; ++I) { 1324 Result.push_back(I); 1325 } 1326 1327 ++SeqPos; 1328 } 1329 1330 return true; 1331 } 1332 1333 Optional<StringRef> DWARFDebugLine::LineTable::getSourceByIndex(uint64_t FileIndex, 1334 FileLineInfoKind Kind) const { 1335 if (Kind == FileLineInfoKind::None || !Prologue.hasFileAtIndex(FileIndex)) 1336 return None; 1337 const FileNameEntry &Entry = Prologue.getFileNameEntry(FileIndex); 1338 if (auto E = dwarf::toString(Entry.Source)) 1339 return StringRef(*E); 1340 return None; 1341 } 1342 1343 static bool isPathAbsoluteOnWindowsOrPosix(const Twine &Path) { 1344 // Debug info can contain paths from any OS, not necessarily 1345 // an OS we're currently running on. Moreover different compilation units can 1346 // be compiled on different operating systems and linked together later. 1347 return sys::path::is_absolute(Path, sys::path::Style::posix) || 1348 sys::path::is_absolute(Path, sys::path::Style::windows); 1349 } 1350 1351 bool DWARFDebugLine::Prologue::getFileNameByIndex( 1352 uint64_t FileIndex, StringRef CompDir, FileLineInfoKind Kind, 1353 std::string &Result, sys::path::Style Style) const { 1354 if (Kind == FileLineInfoKind::None || !hasFileAtIndex(FileIndex)) 1355 return false; 1356 const FileNameEntry &Entry = getFileNameEntry(FileIndex); 1357 auto E = dwarf::toString(Entry.Name); 1358 if (!E) 1359 return false; 1360 StringRef FileName = *E; 1361 if (Kind == FileLineInfoKind::RawValue || 1362 isPathAbsoluteOnWindowsOrPosix(FileName)) { 1363 Result = std::string(FileName); 1364 return true; 1365 } 1366 if (Kind == FileLineInfoKind::BaseNameOnly) { 1367 Result = std::string(llvm::sys::path::filename(FileName)); 1368 return true; 1369 } 1370 1371 SmallString<16> FilePath; 1372 StringRef IncludeDir; 1373 // Be defensive about the contents of Entry. 1374 if (getVersion() >= 5) { 1375 // DirIdx 0 is the compilation directory, so don't include it for 1376 // relative names. 1377 if ((Entry.DirIdx != 0 || Kind != FileLineInfoKind::RelativeFilePath) && 1378 Entry.DirIdx < IncludeDirectories.size()) 1379 IncludeDir = dwarf::toStringRef(IncludeDirectories[Entry.DirIdx]); 1380 } else { 1381 if (0 < Entry.DirIdx && Entry.DirIdx <= IncludeDirectories.size()) 1382 IncludeDir = dwarf::toStringRef(IncludeDirectories[Entry.DirIdx - 1]); 1383 } 1384 1385 // For absolute paths only, include the compilation directory of compile unit. 1386 // We know that FileName is not absolute, the only way to have an absolute 1387 // path at this point would be if IncludeDir is absolute. 1388 if (Kind == FileLineInfoKind::AbsoluteFilePath && !CompDir.empty() && 1389 !isPathAbsoluteOnWindowsOrPosix(IncludeDir)) 1390 sys::path::append(FilePath, Style, CompDir); 1391 1392 assert((Kind == FileLineInfoKind::AbsoluteFilePath || 1393 Kind == FileLineInfoKind::RelativeFilePath) && 1394 "invalid FileLineInfo Kind"); 1395 1396 // sys::path::append skips empty strings. 1397 sys::path::append(FilePath, Style, IncludeDir, FileName); 1398 Result = std::string(FilePath.str()); 1399 return true; 1400 } 1401 1402 bool DWARFDebugLine::LineTable::getFileLineInfoForAddress( 1403 object::SectionedAddress Address, const char *CompDir, 1404 FileLineInfoKind Kind, DILineInfo &Result) const { 1405 // Get the index of row we're looking for in the line table. 1406 uint32_t RowIndex = lookupAddress(Address); 1407 if (RowIndex == -1U) 1408 return false; 1409 // Take file number and line/column from the row. 1410 const auto &Row = Rows[RowIndex]; 1411 if (!getFileNameByIndex(Row.File, CompDir, Kind, Result.FileName)) 1412 return false; 1413 Result.Line = Row.Line; 1414 Result.Column = Row.Column; 1415 Result.Discriminator = Row.Discriminator; 1416 Result.Source = getSourceByIndex(Row.File, Kind); 1417 return true; 1418 } 1419 1420 // We want to supply the Unit associated with a .debug_line[.dwo] table when 1421 // we dump it, if possible, but still dump the table even if there isn't a Unit. 1422 // Therefore, collect up handles on all the Units that point into the 1423 // line-table section. 1424 static DWARFDebugLine::SectionParser::LineToUnitMap 1425 buildLineToUnitMap(DWARFUnitVector::iterator_range Units) { 1426 DWARFDebugLine::SectionParser::LineToUnitMap LineToUnit; 1427 for (const auto &U : Units) 1428 if (auto CUDIE = U->getUnitDIE()) 1429 if (auto StmtOffset = toSectionOffset(CUDIE.find(DW_AT_stmt_list))) 1430 LineToUnit.insert(std::make_pair(*StmtOffset, &*U)); 1431 return LineToUnit; 1432 } 1433 1434 DWARFDebugLine::SectionParser::SectionParser( 1435 DWARFDataExtractor &Data, const DWARFContext &C, 1436 DWARFUnitVector::iterator_range Units) 1437 : DebugLineData(Data), Context(C) { 1438 LineToUnit = buildLineToUnitMap(Units); 1439 if (!DebugLineData.isValidOffset(Offset)) 1440 Done = true; 1441 } 1442 1443 bool DWARFDebugLine::Prologue::totalLengthIsValid() const { 1444 return TotalLength != 0u; 1445 } 1446 1447 DWARFDebugLine::LineTable DWARFDebugLine::SectionParser::parseNext( 1448 function_ref<void(Error)> RecoverableErrorHandler, 1449 function_ref<void(Error)> UnrecoverableErrorHandler, raw_ostream *OS, 1450 bool Verbose) { 1451 assert(DebugLineData.isValidOffset(Offset) && 1452 "parsing should have terminated"); 1453 DWARFUnit *U = prepareToParse(Offset); 1454 uint64_t OldOffset = Offset; 1455 LineTable LT; 1456 if (Error Err = LT.parse(DebugLineData, &Offset, Context, U, 1457 RecoverableErrorHandler, OS, Verbose)) 1458 UnrecoverableErrorHandler(std::move(Err)); 1459 moveToNextTable(OldOffset, LT.Prologue); 1460 return LT; 1461 } 1462 1463 void DWARFDebugLine::SectionParser::skip( 1464 function_ref<void(Error)> RecoverableErrorHandler, 1465 function_ref<void(Error)> UnrecoverableErrorHandler) { 1466 assert(DebugLineData.isValidOffset(Offset) && 1467 "parsing should have terminated"); 1468 DWARFUnit *U = prepareToParse(Offset); 1469 uint64_t OldOffset = Offset; 1470 LineTable LT; 1471 if (Error Err = LT.Prologue.parse(DebugLineData, &Offset, 1472 RecoverableErrorHandler, Context, U)) 1473 UnrecoverableErrorHandler(std::move(Err)); 1474 moveToNextTable(OldOffset, LT.Prologue); 1475 } 1476 1477 DWARFUnit *DWARFDebugLine::SectionParser::prepareToParse(uint64_t Offset) { 1478 DWARFUnit *U = nullptr; 1479 auto It = LineToUnit.find(Offset); 1480 if (It != LineToUnit.end()) 1481 U = It->second; 1482 DebugLineData.setAddressSize(U ? U->getAddressByteSize() : 0); 1483 return U; 1484 } 1485 1486 void DWARFDebugLine::SectionParser::moveToNextTable(uint64_t OldOffset, 1487 const Prologue &P) { 1488 // If the length field is not valid, we don't know where the next table is, so 1489 // cannot continue to parse. Mark the parser as done, and leave the Offset 1490 // value as it currently is. This will be the end of the bad length field. 1491 if (!P.totalLengthIsValid()) { 1492 Done = true; 1493 return; 1494 } 1495 1496 Offset = OldOffset + P.TotalLength + P.sizeofTotalLength(); 1497 if (!DebugLineData.isValidOffset(Offset)) { 1498 Done = true; 1499 } 1500 } 1501